Claude Code release notes: 2.1.181 moves config into the prompt

Claude Code 2.1.181 is not a big model launch. That makes it more useful for teams that already let a coding agent do real work: more settings can be steered inside the work flow, more interruptions recover without breaking the session, and more agent states are visible when something gets stuck.

A coding agent is an AI assistant that can read code, edit files, use tools and sometimes run commands inside human boundaries. Once it becomes part of daily work, the controls around the run matter as much as the model.

Claude Code release notes: 2.1.181 moves more control into the workflow

The clearest change is the new /config key=value syntax. Anthropic says it works in interactive mode, with -p, and in Remote Control. That makes settings easier to discuss in the same work brief as the task, instead of treating them as something hidden away in a separate config file.

The same release adds CLAUDE_CLIENT_PRESENCE_FILE, an environment variable that can point to a marker file to suppress mobile push notifications while the user is at the machine. For teams using Claude in both the terminal and on mobile, that is a small but useful reduction in noise.

Source: Claude Code v2.1.181 on GitHub, Claude Code CHANGELOG.md, and the npm registry entry for @anthropic-ai/claude-code.

Why it matters for practical Claude workflows

Release 2.1.181 is mostly about friction. Long paragraphs now appear line by line. API connection drops during a thinking phase now retry automatically. The subagent panel is calmer because idle subagents auto-hide after 30 seconds and the list caps at five rows with clearer scroll hints.

MCP, the Model Context Protocol, is a way to connect Claude to external tools and data sources through a standard interface. In this release, the MCP OAuth browser page now matches Claude Code's visual style and can close itself after a successful login. More important for operations: claude mcp get and claude mcp list should no longer show ✓ Connected when tools/list fails. They now show that the tool fetch failed.

That is what makes an agent environment more usable for a Nordic team. Not because everything becomes automated overnight, but because there are fewer silent misunderstandings. Pair this with scoped API keys, secrets stored in a secret manager or environment variables, approval-gated edits, redaction of sensitive excerpts, and a simple run log. Then Claude can do more real work without losing traceability.

Source: Claude Code v2.1.181 release notes.

Try this prompt this week

Human step: Update or check Claude Code so you are running 2.1.181 or later. Then open a project where you already have a recurring agent task, such as bug fixing, documentation cleanup, or test analysis.

Paste this into Claude Code:

Read the project's current settings, MCP configuration, and latest run notes.
Find one Claude Code setting that should be explicit for this task.
If a /config line would help, propose the exact line and explain why. Do not change anything without my approval.
Point out where file edits, retries, or subagents need human review.
End with a five-line run log I can paste into the issue.

Good output should include:

• one concrete setting, or a clear note that no setting change is needed
• a short risk point tied to file writes, tool access, or subagents
• a run log that someone outside the session can understand
• a next step a human can approve or stop

The Hammer angle

For Hammer Automation, this release is a reminder that agent work needs routines, not just better prompts. Tool Forge often starts here: which settings should be visible, which tools may be used, which secrets should never appear in plain text, and what log needs to exist afterward?

Start small. Take one recurring Claude Code flow and make the control layer visible before you scale it.